Repository navigation
test: DO NOT MERGE - CI verification only - subgraph preview identity reproductions for #14569 - #16654
test: DO NOT MERGE - CI verification only - subgraph preview identity reproductions for #14569#16654jaeone94 wants to merge 255 commits into
Conversation
Admitting committed nodeOutputs alongside nodePreviewImages made demoting an exposure a permanent no-op (the ambient rollup kept re-rendering what removeExposure just deleted, and queueStore repopulates nodeOutputs on history load so it survived reload), stacked a preview on the host for every interior output node after every run, and fed committed-output nodes into buildImageUrls, which appends a fresh &rand= per call and re-fetches at latent-preview frame rate. Narrow the gate to nodePreviewImages only. isInputPreviewOutput was exported from nodeOutputStore solely for the dropped branch and has no other external callers, so drop the export too - nodeOutputStore.ts's diff against main is now empty. Also correct two docblock claims: previews render when the host is expanded (v-else-if="!isCollapsed" in LGraphNode.vue), not collapsed, and this composable bails at depth 1 instead of recursing like nodeLocationProgressStates.
exposedNodeIds was built from promotedPreviews[].sourceNodeId, which is the chain-resolved leaf id for exposures that come from a nested subgraph. useAmbientSubgraphPreviews emits immediate-interior ids instead. Since last_node_id is per-LGraph, ids restart at 1 in every subgraph, so a nested exposure resolving to leaf id N could collide with and suppress the host's own interior node N. Build the exclusion set from previewExposureStore.getExposures( rootGraphId, hostLocator) instead - those are, by schema, direct children of the host's own subgraph, so both sides of the dedup are in the same id space. Also extract the promoted/ambient merge into a pure mergeSubgraphPreviews function with unit tests covering no overlap, an overlapping exposure winning over its ambient duplicate, and the namespace-collision regression above.
getNodeImageUrls is an unstubbed vi.fn(), reset to undefined by vi.resetAllMocks(), so `if (!urls?.length) return []` short-circuited before the isDetached bail, the nested-SubgraphNode skip, or the hideOutputImages guard ever ran - deleting any of the three still left every test green. Add mockReturnValue stubs so the detached and nested-SubgraphNode tests actually reach their guard, and add a test for the previously uncovered hideOutputImages guard. Also dedupe the byte-identical "no output" test by differentiating it into a test for the getNodeImageUrls-returns-nothing case, add a regression test locking in the nodeOutputs exclusion from the previous commit, and assert media type via getPreviewMediaType instead of a hardcoded 'image' literal so a change to that derivation isn't invisible to the suite.
A prior comment-cleanup pass reworded these it.fails/test.fail markers to drop the incident name and exact guard line. Without that pointer, whoever eventually fixes LGraphNode.vue:172's !lgraphNode?.isSubgraphNode() guard hits a bare "Error: Expect test to fail" with no context. Add the breadcrumb back.
Adds e2e coverage for pathways the two existing preview specs don't reach: an ambient preview nested two subgraph levels deep bubbles to its immediate host only, not the outermost one; a committed (non-live) interior output never gets conflated with the ambient rollup; and un-promoting a still-live sampler's exposed preview falls back to its ambient preview instead of disappearing. Also fills a small unit gap: useAmbientSubgraphPreviews derives video/audio media types the same way the promoted path already does.
removeExposure had no tombstone, so useAmbientSubgraphPreviews would re-surface a preview the next frame after a user explicitly demoted it, as long as the interior node kept producing live output. Add a suppressedAmbientNodeIds tombstone set to previewExposureStore, keyed per (rootGraphId, hostNodeLocator): removeExposure adds the removed source node to it, addExposure clears it, and clearGraph drops it with the rest of the host's state. useAmbientSubgraphPreviews now excludes suppressed node ids from its interior-node scan. Also documents two invariants raised in review: nodePreviewImages is only ever written by live execution/streaming frames, never static input data, so the single-branch gate needs no isInputPreviewOutput guard; and subgraph.nodes is a plain non-reactive array, so structural add/remove while collapsed isn't tracked until the next preview frame.
… own regression test b6bc646 added a suppressedAmbientNodeIds tombstone so removeExposure would permanently hide an interior node's ambient preview. That directly reverses this PR's stated design ("any interior node currently producing output shows on the host, whether or not anyone promoted it") and broke the e2e regression test added earlier in this branch (subgraphConcurrentPreviews.spec.ts: "un-promoting a still-live sampler falls back to its ambient preview instead of disappearing"), which was failing in CI. Un-promoting a node only removes its explicit exposure (name/ordering in the promoted slot); it was never meant to blacklist the node from the ambient rollup while it keeps producing live output. Revert the tombstone mechanism and its tests, restoring the pre-b6bc646 behavior that the existing regression test expects.
…nitions NodeLocatorId is intentionally definition-scoped, so two SubgraphNode instances of the same definition collapsed to the same preview storage key once production writes went through executionIdToNodeLocatorId(). The later frame overwrote the earlier one and both hosts rendered it. Store live preview frames by their own NodeExecutionId (instance-scoped) alongside the existing NodeLocatorId-keyed map, and read ambient previews through that execution-id path instead of the shared locator.
…-preview-outputs # Conflicts: # src/renderer/extensions/vueNodes/components/LGraphNode.test.ts
…-preview-outputs # Conflicts: # src/stores/nodeOutputStore.ts
Six e2e scenarios plus one unit repro covering preview identity behaviour on top of #14569. Three e2e scenarios and the unit repro are expected to fail: they pin the current behaviour for review, not a desired state. Verification only — this branch is not intended to merge.
|
Important Review skippedAuto reviews are disabled on base/target branches other than the default branch. Please check the settings in the CodeRabbit UI or the ⚙️ Run configurationConfiguration used: Path: .coderabbit.yaml Review profile: ASSERTIVE Plan: Advanced Run ID: You can disable this status message by setting the Use the checkbox below for a quick retry:
Comment |
🎭 Playwright: ❌ 1946 passed, 1 failed · 2 flaky❌ Failed Tests📊 Browser Reports
🎨 Storybook: ✅ Built — View Storybook📦 Bundle Size
⚡ Performance Report
Absolute values
Raw data{
"timestamp": "2026-09-09T19:08:54.089Z",
"gitSha": "a04f9515d9d1f017810e2c5b0a914a73106bdcb5",
"branch": "jaeone/verify/subgraph-preview-identity-e2e",
"measurements": [
{
"name": "canvas-idle",
"durationMs": 2006.8989999999758,
"styleRecalcs": 9,
"styleRecalcDurationMs": 6.501000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 367.244,
"heapDeltaBytes": -2428148,
"heapUsedBytes": 59457148,
"domNodes": -261,
"jsHeapTotalBytes": 4186112,
"scriptDurationMs": 7.763000000000003,
"eventListeners": -154,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "canvas-idle",
"durationMs": 1990.2850000000285,
"styleRecalcs": 9,
"styleRecalcDurationMs": 5.882,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 365.182,
"heapDeltaBytes": 14982004,
"heapUsedBytes": 76927052,
"domNodes": -260,
"jsHeapTotalBytes": 4710400,
"scriptDurationMs": 7.254999999999997,
"eventListeners": -154,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "canvas-mouse-sweep",
"durationMs": 1806.572000000017,
"styleRecalcs": 74,
"styleRecalcDurationMs": 23.858000000000004,
"layouts": 14,
"layoutDurationMs": 4.041,
"taskDurationMs": 738.2990000000001,
"heapDeltaBytes": 16623584,
"heapUsedBytes": 78351040,
"domNodes": -262,
"jsHeapTotalBytes": 5234688,
"scriptDurationMs": 141.97600000000003,
"eventListeners": -154,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.699999999999818
},
{
"name": "canvas-mouse-sweep",
"durationMs": 1727.630999999974,
"styleRecalcs": 74,
"styleRecalcDurationMs": 22.822999999999997,
"layouts": 14,
"layoutDurationMs": 3.698,
"taskDurationMs": 677.0649999999999,
"heapDeltaBytes": 17785840,
"heapUsedBytes": 79588492,
"domNodes": -261,
"jsHeapTotalBytes": 4710400,
"scriptDurationMs": 123.79399999999998,
"eventListeners": -152,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.699999999999818
},
{
"name": "canvas-zoom-sweep",
"durationMs": 1710.040000000049,
"styleRecalcs": 31,
"styleRecalcDurationMs": 12.150999999999998,
"layouts": 6,
"layoutDurationMs": 1.0260000000000002,
"taskDurationMs": 268.07399999999996,
"heapDeltaBytes": 2179148,
"heapUsedBytes": 64169796,
"domNodes": 75,
"jsHeapTotalBytes": 4980736,
"scriptDurationMs": 9.704,
"eventListeners": 21,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.699999999999818
},
{
"name": "canvas-zoom-sweep",
"durationMs": 1707.3390000000472,
"styleRecalcs": 30,
"styleRecalcDurationMs": 11.641,
"layouts": 6,
"layoutDurationMs": 0.9140000000000001,
"taskDurationMs": 261.231,
"heapDeltaBytes": 2201692,
"heapUsedBytes": 64164284,
"domNodes": 75,
"jsHeapTotalBytes": 4718592,
"scriptDurationMs": 9.523,
"eventListeners": 21,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "dom-widget-clipping",
"durationMs": 526.0379999999714,
"styleRecalcs": 10,
"styleRecalcDurationMs": 5.949,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 329.698,
"heapDeltaBytes": 17878584,
"heapUsedBytes": 79941488,
"domNodes": 16,
"jsHeapTotalBytes": 5505024,
"scriptDurationMs": 133.493,
"eventListeners": 2,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "dom-widget-clipping",
"durationMs": 533.504999999991,
"styleRecalcs": 12,
"styleRecalcDurationMs": 7.305000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 355.115,
"heapDeltaBytes": 17481928,
"heapUsedBytes": 79539256,
"domNodes": 20,
"jsHeapTotalBytes": 5242880,
"scriptDurationMs": 134.062,
"eventListeners": 2,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333335,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-idle",
"durationMs": 2020.545000000027,
"styleRecalcs": 7,
"styleRecalcDurationMs": 4.955000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 457.39400000000006,
"heapDeltaBytes": 3314520,
"heapUsedBytes": 82858336,
"domNodes": -245,
"jsHeapTotalBytes": -1318912,
"scriptDurationMs": 27.663,
"eventListeners": -180,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.699999999999818
},
{
"name": "large-graph-idle",
"durationMs": 2024.107000000015,
"styleRecalcs": 9,
"styleRecalcDurationMs": 6.737,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 490.8340000000001,
"heapDeltaBytes": 226524,
"heapUsedBytes": 80170300,
"domNodes": -242,
"jsHeapTotalBytes": -1318912,
"scriptDurationMs": 29.196999999999992,
"eventListeners": -180,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333335,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-pan",
"durationMs": 2162.7450000000294,
"styleRecalcs": 69,
"styleRecalcDurationMs": 13.314,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 1322.375,
"heapDeltaBytes": -2057692,
"heapUsedBytes": 78195656,
"domNodes": -248,
"jsHeapTotalBytes": 1265664,
"scriptDurationMs": 637.134,
"eventListeners": -180,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-pan",
"durationMs": 2115.877000000012,
"styleRecalcs": 67,
"styleRecalcDurationMs": 10.914,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 1248.476,
"heapDeltaBytes": 14044620,
"heapUsedBytes": 78567912,
"domNodes": 10,
"jsHeapTotalBytes": 1011712,
"scriptDurationMs": 631.596,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "large-graph-zoom",
"durationMs": 3194.5349999999735,
"styleRecalcs": 64,
"styleRecalcDurationMs": 13.011999999999999,
"layouts": 60,
"layoutDurationMs": 11.238000000000001,
"taskDurationMs": 1434.1720000000003,
"heapDeltaBytes": -9344640,
"heapUsedBytes": 73824840,
"domNodes": -251,
"jsHeapTotalBytes": -1019904,
"scriptDurationMs": 652.038,
"eventListeners": -178,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "large-graph-zoom",
"durationMs": 3155.463999999938,
"styleRecalcs": 65,
"styleRecalcDurationMs": 13.572000000000001,
"layouts": 60,
"layoutDurationMs": 10.899,
"taskDurationMs": 1390.4840000000002,
"heapDeltaBytes": -12941924,
"heapUsedBytes": 69106688,
"domNodes": -252,
"jsHeapTotalBytes": -532480,
"scriptDurationMs": 628.563,
"eventListeners": -178,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "legacy-node-drag",
"durationMs": 2380.5659999999875,
"styleRecalcs": 42,
"styleRecalcDurationMs": 5.97,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 1564.171,
"heapDeltaBytes": 14432952,
"heapUsedBytes": 80517796,
"domNodes": 4,
"jsHeapTotalBytes": 1204224,
"scriptDurationMs": 787.8040000000001,
"eventListeners": 186,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "legacy-node-drag",
"durationMs": 2228.7549999999783,
"styleRecalcs": 43,
"styleRecalcDurationMs": 6.187999999999999,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 1488.3999999999999,
"heapDeltaBytes": 5415164,
"heapUsedBytes": 70885740,
"domNodes": 6,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 755.631,
"eventListeners": 186,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.699999999999818
},
{
"name": "minimap-idle",
"durationMs": 2034.4650000000115,
"styleRecalcs": 7,
"styleRecalcDurationMs": 5.619000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 484.544,
"heapDeltaBytes": -8423160,
"heapUsedBytes": 78783016,
"domNodes": -251,
"jsHeapTotalBytes": 552960,
"scriptDurationMs": 29.597000000000005,
"eventListeners": -180,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "minimap-idle",
"durationMs": 2006.6080000000284,
"styleRecalcs": 7,
"styleRecalcDurationMs": 5.728000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 467.99099999999993,
"heapDeltaBytes": -5335948,
"heapUsedBytes": 82094380,
"domNodes": -246,
"jsHeapTotalBytes": -1019904,
"scriptDurationMs": 28.553,
"eventListeners": -180,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333335,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "subgraph-dom-widget-clipping",
"durationMs": 496.2749999999687,
"styleRecalcs": 46,
"styleRecalcDurationMs": 8.690000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 310.97799999999995,
"heapDeltaBytes": 13398664,
"heapUsedBytes": 76697364,
"domNodes": 18,
"jsHeapTotalBytes": 5767168,
"scriptDurationMs": 114.08000000000001,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.799999999999727
},
{
"name": "subgraph-dom-widget-clipping",
"durationMs": 508.0039999999144,
"styleRecalcs": 47,
"styleRecalcDurationMs": 8.838000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 316.458,
"heapDeltaBytes": 13485896,
"heapUsedBytes": 76712576,
"domNodes": 20,
"jsHeapTotalBytes": 5505024,
"scriptDurationMs": 112.787,
"eventListeners": 6,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66666666666665,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-idle",
"durationMs": 2010.3260000000205,
"styleRecalcs": 10,
"styleRecalcDurationMs": 6.734,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 387.583,
"heapDeltaBytes": 12215412,
"heapUsedBytes": 75121768,
"domNodes": -276,
"jsHeapTotalBytes": 4710400,
"scriptDurationMs": 7.786,
"eventListeners": -198,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "subgraph-idle",
"durationMs": 1991.755000000012,
"styleRecalcs": 10,
"styleRecalcDurationMs": 6.057000000000001,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 289.647,
"heapDeltaBytes": -3857540,
"heapUsedBytes": 59505452,
"domNodes": -275,
"jsHeapTotalBytes": 5234688,
"scriptDurationMs": 6.588,
"eventListeners": -230,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.66333333333332,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-mouse-sweep",
"durationMs": 1722.2710000000347,
"styleRecalcs": 78,
"styleRecalcDurationMs": 23.706,
"layouts": 18,
"layoutDurationMs": 4.822,
"taskDurationMs": 587.583,
"heapDeltaBytes": -1088488,
"heapUsedBytes": 61802228,
"domNodes": -273,
"jsHeapTotalBytes": 4448256,
"scriptDurationMs": 89.857,
"eventListeners": -200,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.800000000000182
},
{
"name": "subgraph-mouse-sweep",
"durationMs": 1677.5619999999662,
"styleRecalcs": 77,
"styleRecalcDurationMs": 21.676,
"layouts": 18,
"layoutDurationMs": 4.414000000000001,
"taskDurationMs": 543.03,
"heapDeltaBytes": -1870484,
"heapUsedBytes": 60650892,
"domNodes": -277,
"jsHeapTotalBytes": 4972544,
"scriptDurationMs": 83.539,
"eventListeners": -228,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "subgraph-transition-enter",
"durationMs": 866.8880000000172,
"styleRecalcs": 18,
"styleRecalcDurationMs": 22.823000000000004,
"layouts": 5,
"layoutDurationMs": 9.235,
"taskDurationMs": 688.1440000000001,
"heapDeltaBytes": 19626232,
"heapUsedBytes": 100755772,
"domNodes": 13587,
"jsHeapTotalBytes": 10747904,
"scriptDurationMs": 21.318000000000005,
"eventListeners": 2385,
"totalBlockingTimeMs": 98,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "viewport-pan-sweep",
"durationMs": 8167.292000000032,
"styleRecalcs": 249,
"styleRecalcDurationMs": 37.394999999999996,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 4483.147,
"heapDeltaBytes": -12710752,
"heapUsedBytes": 66454612,
"domNodes": -243,
"jsHeapTotalBytes": 4448256,
"scriptDurationMs": 2117.8360000000002,
"eventListeners": -136,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.670000000000012,
"p95FrameDurationMs": 16.80000000000109
},
{
"name": "viewport-pan-sweep",
"durationMs": 8270.052999999962,
"styleRecalcs": 250,
"styleRecalcDurationMs": 36.223,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 4482.685,
"heapDeltaBytes": -6097176,
"heapUsedBytes": 74622404,
"domNodes": -242,
"jsHeapTotalBytes": -1318912,
"scriptDurationMs": 2109.908,
"eventListeners": -166,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "vue-large-graph-idle",
"durationMs": 14392.560000000003,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 13683.318999999998,
"heapDeltaBytes": -39168292,
"heapUsedBytes": 175757860,
"domNodes": -6542,
"jsHeapTotalBytes": 802816,
"scriptDurationMs": 213.19699999999997,
"eventListeners": -7478,
"totalBlockingTimeMs": 0,
"frameDurationMs": 17.779999999999927,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "vue-large-graph-idle",
"durationMs": 14331.041000000027,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 13785.909,
"heapDeltaBytes": -18785136,
"heapUsedBytes": 195078900,
"domNodes": -6542,
"jsHeapTotalBytes": 3158016,
"scriptDurationMs": 210.194,
"eventListeners": -7478,
"totalBlockingTimeMs": 0,
"frameDurationMs": 17.219999999999953,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "vue-large-graph-pan",
"durationMs": 17096.30599999997,
"styleRecalcs": 142,
"styleRecalcDurationMs": 14.431,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 16593.733999999997,
"heapDeltaBytes": -9400400,
"heapUsedBytes": 188903176,
"domNodes": -6542,
"jsHeapTotalBytes": -2150400,
"scriptDurationMs": 747.731,
"eventListeners": -7476,
"totalBlockingTimeMs": 0,
"frameDurationMs": 17.223333333333358,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "vue-large-graph-pan",
"durationMs": 17102.543999999965,
"styleRecalcs": 137,
"styleRecalcDurationMs": 14.629000000000003,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 16741.449000000004,
"heapDeltaBytes": -15861048,
"heapUsedBytes": 185768576,
"domNodes": -6542,
"jsHeapTotalBytes": 208896,
"scriptDurationMs": 740.418,
"eventListeners": -7476,
"totalBlockingTimeMs": 15,
"frameDurationMs": 17.776666666666642,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "vue-node-resize-workload",
"durationMs": 5070.432000000039,
"styleRecalcs": 20,
"styleRecalcDurationMs": 61.53699999999998,
"layouts": 20,
"layoutDurationMs": 21.41,
"taskDurationMs": 4985.939,
"heapDeltaBytes": -36823796,
"heapUsedBytes": 179785536,
"domNodes": -6542,
"jsHeapTotalBytes": 19107840,
"scriptDurationMs": 84.10900000000001,
"eventListeners": -7476,
"totalBlockingTimeMs": 245,
"frameDurationMs": 17.223333333333297,
"p95FrameDurationMs": 16.80000000000109
},
{
"name": "vue-node-resize-workload",
"durationMs": 5049.576999999999,
"styleRecalcs": 20,
"styleRecalcDurationMs": 62.198,
"layouts": 20,
"layoutDurationMs": 21.73699999999999,
"taskDurationMs": 4984.727999999999,
"heapDeltaBytes": -34930512,
"heapUsedBytes": 167583644,
"domNodes": -6542,
"jsHeapTotalBytes": 13385728,
"scriptDurationMs": 83.54499999999999,
"eventListeners": -7478,
"totalBlockingTimeMs": 279,
"frameDurationMs": 17.779999999999987,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "vue-zoom-out-idle",
"durationMs": 7206.489999999974,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 6914.415000000002,
"heapDeltaBytes": 17744200,
"heapUsedBytes": 181297596,
"domNodes": -1,
"jsHeapTotalBytes": 0,
"scriptDurationMs": 105.24699999999999,
"eventListeners": -10,
"totalBlockingTimeMs": 0,
"frameDurationMs": 17.223333333333358,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "vue-zoom-out-idle",
"durationMs": 7202.777999999967,
"styleRecalcs": 0,
"styleRecalcDurationMs": 0,
"layouts": 0,
"layoutDurationMs": 0,
"taskDurationMs": 6917.243999999999,
"heapDeltaBytes": 1634640,
"heapUsedBytes": 185774804,
"domNodes": 0,
"jsHeapTotalBytes": -7602176,
"scriptDurationMs": 104.852,
"eventListeners": -10,
"totalBlockingTimeMs": 0,
"frameDurationMs": 17.220000000000073,
"p95FrameDurationMs": 16.799999999999272
},
{
"name": "workflow-execution",
"durationMs": 423.19700000001603,
"styleRecalcs": 13,
"styleRecalcDurationMs": 21.165999999999997,
"layouts": 4,
"layoutDurationMs": 1.777,
"taskDurationMs": 100.70700000000001,
"heapDeltaBytes": 4940636,
"heapUsedBytes": 66663172,
"domNodes": 122,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 11.595,
"eventListeners": 65,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000728
},
{
"name": "workflow-execution",
"durationMs": 446.5490000000045,
"styleRecalcs": 18,
"styleRecalcDurationMs": 18.802,
"layouts": 4,
"layoutDurationMs": 1.8120000000000003,
"taskDurationMs": 94.408,
"heapDeltaBytes": 4967252,
"heapUsedBytes": 66647436,
"domNodes": 146,
"jsHeapTotalBytes": 262144,
"scriptDurationMs": 8.123999999999999,
"eventListeners": 83,
"totalBlockingTimeMs": 0,
"frameDurationMs": 16.666666666666668,
"p95FrameDurationMs": 16.700000000000273
}
]
}🔬 E2E Coverage
Top 10 files by uncovered lines
🌐 Website E2ETip All tests passed.
🔗 Website PreviewWebsite Preview: https://comfy-website-preview-pr-16654.vercel.app Matches production — what ships if we release right now. This commit: https://website-frontend-arsqewuwz-comfyui.vercel.app Last updated: 2026-09-09T19:00:06Z for |
Loading a workflow with two subgraph definitions that collide on an interior node id renumbers the colliding node via deduplicateSubgraphNodeIds, but properties.previewExposures on the host was never patched to follow — same as patchProxyWidgets already does for legacy proxyWidgets. The stale sourceNodeId then resolves to nothing, so the promoted preview silently disappears after reload. Adds patchPreviewExposures, applied alongside patchProxyWidgets for both interior and root-level hosts, plus a regression test. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BPhqLweRXNznT3YWP9Mx5
useAmbientSubgraphPreviews built its interior-node lookup key from the host's own bare local id (e.g. "10"), not its root-relative execution id (e.g. "9:10" for a host nested one level deep). For any host that is itself nested inside another subgraph, that means the ambient rollup never finds the frames app.ts's ancestor fan-out already wrote for it, so a live interior preview never renders on a nested host. Reuses the existing getExecutionIdByNode graph-traversal helper (already used the same way by error reporting, missing-node/model scanning, etc.) instead of the bare-id normalization, plus a regression test with a host nested inside another subgraph. Co-Authored-By: Claude Sonnet 5 <noreply@anthropic.com> Claude-Session: https://claude.ai/code/session_011BPhqLweRXNznT3YWP9Mx5
Cross-workflow anonymous operation results are ignored. The active workflow still accepts its own anonymous result. Focused CRDT tests and typecheck are green. <details><summary>Full context for agent readers</summary> A late `doc_ops_result` carries its source workflow ID, but the follower adapter previously stripped it before passing the result to `createOpSender`. An anonymous result from workflow A could therefore acknowledge workflow B's in-flight batch. This change forwards `workflowId` through `OpsResultView` and drops an explicitly mismatched result before operation-ID or anonymous settlement. Results without a workflow ID retain the existing fallback behavior. Regression coverage includes the sender boundary and the full A-to-B workflow switch path. Source review: #16655 (comment) Verification: - `pnpm test:unit src/workbench/extensions/agent/crdt/opSender.test.ts src/workbench/extensions/agent/crdt/crossWorkflowPending.test.ts` — 28 passed - `pnpm typecheck` — passed - scoped ESLint and oxfmt — passed - pre-push Knip — passed </details> --------- Co-authored-by: Christian Byrne <christian-byrne@users.noreply.github.com>
… answer (#17180) *PR Created by the Glary-Bot Agent* --- Updates the Q&A on [comfy.org/minimax/license](https://comfy.org/minimax/license) with copy supplied by GTM in #gtm. ## Changes **1. Replaced the answer to "Who needs a MiniMax H3 commercial license?"** — now covers the H3 Community License $20M revenue threshold and the four excluded territories, ending with a link to `platform.minimax.io/h3-license`. **2. Added a new Q&A on pricing parity**, placed after the Professional/Enterprise tier question: > **Is the Commercial License more expensive through Comfy?** > No. **MiniMax requires pricing parity across its Commercial License resellers**, so purchasing the license through Comfy does not come with a reseller markup or a higher license price compared with other authorized channels. **3. Added `**phrase**` emphasis support to FAQ answers.** `parseFaqAnswer` previously understood links only, so the bolded clause needed a new `strong` part type. Both link forms are claimed before emphasis and nothing nests, so a bold span overlapping a link keeps the link and loses its emphasis; delimiters are stripped either way, so no `**` can reach the page or the `FAQPage` structured data. Rendered in both parser consumers (`FAQSplit01`, `ModelLaunchStepsSection`). Both answers are localised (`en` + `zh-CN`). This is the first FAQ answer anywhere to use a paragraph break — it renders via the pre-existing `whitespace-pre-line` on the answer paragraph. ## Copy correction confirmed with GTM The draft supplied in Slack said users in the US/EU/UK/South Korea must acquire a *"community license"* — for the territories that same sentence describes as **excluded** from the Community License, so the answer contradicted itself and never answered its own question. @U08HX6GS2GL and @U045ZEVQA1K confirmed in thread this should read **"commercial license"**, and that correction is applied in both locales. The later reference to what the *Community* License excludes is deliberately unchanged, so the sentence now reads: acquire a **commercial** license for the territories excluded from the **Community** License. ## Remaining copy notes for GTM (non-blocking) 1. **"pricing parity across its Commercial License resellers"** and *"other authorized channels"* read as though other resellers exist, while this page's hero says **"only official reseller"**. These can coexist if "other channels" means MiniMax selling direct (which this same answer now links to) — worth a conscious call either way, since the parity line is a factual claim about a partner's contract terms. 2. `"You can acquire **them** here"` refers back to a singular licence (en only; the zh-CN string is unaffected). This wording also lands verbatim in the structured data. 3. The link's visible anchor text is the raw URL, whereas other links on the page use a label. Left as supplied rather than rewriting your sentence — happy to switch it to a labelled link if you prefer. 4. Worth confirming the licence facts themselves ($20M threshold, the four excluded territories, "separate written authorization") against the licence text, since they now appear on a public page and in structured data. ## Verification - `pnpm test:unit` (website) — 1025 pass; `pnpm typecheck:website` — 0 errors; oxlint/eslint/oxfmt clean on changed files; `astro build` — 724 pages, exit 0 - Screenshots below are from the **production build** served via `astro preview`, both locales - Asserted in the built output: corrected wording present in visible text *and* `FAQPage` JSON-LD, old wording absent, the Community-License exclusion reference intact, bold still renders, no raw `**` anywhere visible - New tests cover bold parsing, bold+link coexistence, bold nested in a link label, bold straddling a link boundary, unclosed markers, pairing across a paragraph break, and marker stripping for structured data Pre-existing and untouched: `cloudNodes.test.ts` has 9 oxlint `no-unnecessary-type-assertion` errors on `main`. Known limitation: the markdown twin (`llms.txt`) collapses the new answer's paragraph break to a space — text is complete, only the break is lost. ## Screenshots   --------- Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
## Summary Follow up on #17010 by addressing its three review findings. ## Changes - **What**: Assert promoted-widget metadata is removed when its source node is deleted, restore the actionable #14495 expected-failure note, and remove the orphaned textarea test ID. ## Review Focus This PR depends on #17010. Its diff will narrow to these follow-up changes when that PR merges. --------- Co-authored-by: Amp <amp@ampcode.com>
*PR Created by the Glary-Bot Agent*
---
Reported in #website-and-docs: on the Yearly toggle,
`comfy.org/cloud/pricing` and the in-app **Choose a Plan** modal both
showed "4,200 monthly credits" / "7,400 monthly credits" / "21,100
monthly credits", so a user read it as a monthly refill.
Annual plans don't refill monthly — they grant the whole year up front.
`stripe_catalog.yaml` sets `credit_grant` to `50400` / `88800` /
`253200` on `standard-annual` / `creator-annual` / `pro-annual` (12× the
monthly grants), team stops use `annual quantity = 12 ×
credits_monthly`, and `docs/support/subscription/subscribing.mdx`
already states "Annual plans receive their full year of credits up
front."
## Changes
Yearly now shows the year's total under a "credits per year" label;
Monthly is unchanged. The video estimates scale with it so the two lines
stay consistent.
| Surface | Yearly (before → after) |
|---|---|
| `/cloud/pricing` plan + team cards | `4,200 monthly credits` → `50,400
credits per year` |
| In-app **Choose a Plan** (`UnifiedPricingTable`) | same, personal
tiers + team card |
| Legacy `PricingTable` | already showed `50,400 / Total yearly
credits`; its video estimate was still the monthly `~380` |
| Legacy team `PricingTableWorkspace` | `Monthly credits / member` →
`Yearly credits / member` |
Both flagged surfaces are separate components in this repo, so each was
fixed; the shared `amountForBillingCycle` helper in `tierPricing.ts`
holds the one ×12 rule.
**Prose on the same page also had to move.** The plan cards said "per
year" while `What's included` and the pricing FAQ underneath still said
"All plans will include a **monthly pool** of credits" — the page
contradicted itself and still asserted the misconception. Updated in
both locales: `pricing.included.feature4`/`feature8`, plus the
`how-does-pricing-work`, `do-credits-roll-over`, `run-out-of-credits`
and `team-upgrade-carryover` FAQ entries.
**Drive-by:** `teamCreditTiers.ts` had `videos: 47830` for the $2,500
stop; every other stop follows `credits/21100 × 1915`, which gives
`47875` — the value the in-app table derives. Multiplying by 12 turned a
45-video typo into a 540-video gap between the two surfaces, so it's
corrected here. Revert that one line if the old figure was deliberate.
## Tests
New coverage on all four tables asserts the label and figure flip with
the cycle. Two invariant tests replace hand-checking the typed-in
strings: yearly credits/estimates must be 12× the monthly ones in every
shipped locale, and every team stop's video count must follow the same
credits-per-video rate. Both were confirmed to fail when a value is
mutated.
`pnpm typecheck`, `lint`, `format:check`, `knip`, `test:unit` (in-app +
website) all pass. Two full-suite unit failures (`presetService`,
`useFirstRunTourController`) reproduce on an unmodified `main` and pass
in isolation — pre-existing flakes.
## Before merge
Please run **`/update-website-screenshots`** (or add the `Update Website
Screenshots` label). The `pricing-tiers-*` and `pricing-faq-*`
Playwright baselines change, and the visual job fails until they're
regenerated. The 8 `pricing-tiers-*`/`pricing-faq-*` baselines are
updated in this commit, but the sandbox doesn't reproduce the CI
container's font rendering (10 of the 62 visual tests already fail there
on an untouched `main` with 530–3811px sub-pixel diffs), so treat the
committed images as provisional and regenerate them in CI via the label
before merge.
Two new `src/locales/en/main.json` keys (`subscription.yearlyCredits`,
`subscription.yearlyCreditsPerMemberLabel`) are pending translation; per
repo convention non-English locales are generated at release, and they
fall back to English until then.
## Follow-up (not in this PR)
`CreditsTile.vue` renders "Monthly credits are used up. Refills {date}"
without branching on subscription duration, so an annual subscriber who
exhausts their up-front grant still meets the same wording on the
balance surface.
## Screenshots





---------
Co-authored-by: Glary-Bot <glary-bot@users.noreply.github.com>
Co-authored-by: github-actions <github-actions@github.com>
Co-authored-by: Deep Mehta <42841935+deepme987@users.noreply.github.com>
Co-authored-by: Deep Mehta <deepme987@gmail.com>
## Summary Migrates legacy `vi.mock` calls in the remaining component tests to the dynamic `import()` form as an independent slice of #16174. ## Changes - **What**: Converts mocks across 95 files; legacy partial mocks use `<unknown>` to preserve the string overload's existing type contract. - **Breaking**: None. ## Review Focus This is a mechanical syntax migration. The lint rule and documentation are intentionally deferred until every domain slice has merged. Verified with the full pre-commit lint/typecheck gates, `pnpm knip`, and the combined unit suite (1,506 files; 20,337 passed). --------- Co-authored-by: Amp <amp@ampcode.com>
## Summary Migrates legacy `vi.mock` calls in shared composable tests to the dynamic `import()` form as an independent slice of #16174. ## Changes - **What**: Converts mocks across 85 files; legacy partial mocks use `<unknown>` to preserve the string overload's existing type contract. - **Breaking**: None. ## Review Focus This is a mechanical syntax migration. The lint rule and documentation are intentionally deferred until every domain slice has merged. Verified with the full pre-commit lint/typecheck gates, `pnpm knip`, and the combined unit suite (1,506 files; 20,337 passed). Co-authored-by: Amp <amp@ampcode.com>
## Summary Migrates legacy `vi.mock` calls in cloud and workspace tests to the dynamic `import()` form as an independent slice of #16174. ## Changes - **What**: Converts mocks across 84 files; legacy partial mocks use `<unknown>` to preserve the string overload's existing type contract. - **Breaking**: None. ## Review Focus This is a mechanical syntax migration. The lint rule and documentation are intentionally deferred until every domain slice has merged. Verified with the full pre-commit lint/typecheck gates, `pnpm knip`, and the combined unit suite (1,506 files; 20,337 passed). Co-authored-by: Amp <amp@ampcode.com>
…Rules (#17094) Addresses @AustinMroz's pluralization comment on #16748. The model grid held the node count as two hardcoded translation keys, `oneNode` and `threeNodes`, containing the literal strings `1 node` and `3 nodes`. That is not internationalization: it bakes English plural rules into key names, and it only works because today's lineup happens to be five ones and a three. A model backing two or four nodes would have needed another key. The count is data now (`nodeCount: number`) and the plural form is selected at render time: ```ts function nodeCountLabel(nodeCount: number): string { const key = new Intl.PluralRules(locale).select(nodeCount) === 'one' ? 'cloudNodesLaunch.models.nodeCountOne' : 'cloudNodesLaunch.models.nodeCountOther' return t(key, locale).replace('{count}', String(nodeCount)) } ``` **On the linked vue-i18n docs:** `vue-i18n` is not a dependency of `apps/website` — this app has its own minimal `t(key, locale)` with no plural or interpolation support, so that syntax is not available here. `Intl.PluralRules` is, and the app already uses exactly this pattern in `components/cloud-nodes/PackCard.vue`, which pluralizes a node count with the same `.one`/`.other` key pair and `{count}` replacement. This follows that rather than introducing a second approach. Worth noting the zh-CN forms are identical, since Chinese has no plural. Going through `Intl.PluralRules` is what makes that correct rather than coincidental. The test got simpler too: it sums `nodeCount` to prove the grid still covers the eight nodes we ship, instead of mapping key names back to numbers via `endsWith`. `copy.test` never covered these two keys, so it does now. **Base:** stacked on `deepme987/website/comfy-cloud-nodes-launch` because `templates/cloud-nodes/` only exists there. GitHub will retarget this to `main` when #16748 merges. Gates: oxlint, eslint, oxfmt, knip, typecheck (0 errors), 1034 tests.
## Summary Fetch complete PR history before lockfile repair workflows attempt to merge the base SHA, preventing shallow clones from being misclassified as unrelated histories. ## Changes - **What**: Set `fetch-depth: 0` on all producer and privileged consumer checkouts that reproduce the lockfile-only merge conflict. ## Review Focus PR #17138's repair run logged `fatal: refusing to merge unrelated histories`, selected repair mode, and pushed nothing. Confirm complete history is available at each merge step without changing the existing artifact and privilege boundaries. Co-authored-by: Amp <amp@ampcode.com>
## Summary Adds the first Workshop page at `/workshop`. It lists the 268 Router models from the committed catalog and lets people search them or filter by provider and output type. The page renders the first 48 model links into the HTML, then loads more on request. Search and filters continue to use the complete catalog. Model parameter schemas are not sent to this page. Workshop remains absent from release builds. In Workshop-enabled labeled previews, the catalog is indexable and appears in the sitemap as an English-only, locale-invariant route. Static `llms.txt` continues to omit it until launch because that file cannot vary by build shape. Model links are implemented in the next slice. ## Test plan - [x] Search, provider, and output filters tested together - [x] Show more pagination tested - [x] Catalog has 268 unique model IDs and links - [x] Astro typecheck passes - [x] Website production build passes - [x] Release build emits no Workshop route or `llms.txt` link - [x] Enabled build includes `/workshop/` in the sitemap with valid hreflang targets - [x] Built page contains 48 initial model links and no parameter schemas - [x] Knip and repository pre-push checks pass
## Summary Require authors and reviewers to investigate type and lint failures instead of accepting suppressions with explanatory comments. ## Changes - **What**: Add a diagnostic-first workflow for suppressions, assertions, typed test fixtures, semantic queries, and deterministic async tests. - **What**: Make blanket lint disables and unsupported type overrides explicit review blockers. - **What**: Update the automated review prompt to reproduce suppressed diagnostics before approval. ## Review Focus Confirm that the exception language stays narrow enough for verified external constraints without making comments a substitute for evidence. --------- Co-authored-by: Amp <amp@ampcode.com>
## Summary Enable `vitest/prefer-import-in-mock` for Oxlint-covered tests now that the split migration is complete. ## Changes - **What**: Migrates tests added during the split rollout, replaces remaining `vue-i18n` module mocks with real plugins, enables the lint rule, and updates testing guidance. ## Review Focus Check that `<unknown>` remains limited to legacy partial factories that cannot satisfy the full module type. --------- Co-authored-by: Amp <amp@ampcode.com>
## Summary Derives the form fields for every Workshop model from the canonical `parameters` and `roles` stored in the packed catalog. Derived fields are not persisted beside their source schema; they are created at the model-detail consumption boundary. The derivation produces text, select, number, toggle, and media controls. It removes the model ID and Router transport settings from the visible form. Complex objects and lists remain editable as JSON text, so models with required complex inputs remain representable. Open string unions retain their curated values as suggestions instead of becoming closed selects. Numeric precision follows `multipleOf`, integers use step 1, and unrestricted numbers use `step="any"`. Boolean defaults and true-only constraints are preserved exactly from the source schema. ## Test plan - [x] Test each derived field type - [x] Test media roles and file types - [x] Test complex inputs represented as JSON text - [x] Test open string unions remain editable - [x] Test numeric steps do not narrow the source schema - [x] Test optional booleans do not acquire defaults - [x] Test required true-only booleans preserve their constraint - [x] Check every required input across the packed catalog - [x] Check every model has form fields - [x] Website unit tests, typecheck, build, E2E, lint, and patch coverage pass
## Summary Remove two LiteGraph partial mocks by importing the real graph-ID resolver from its owning utility module. ## Changes - **What**: Import `resolveNodeRootGraphId` directly from `utils/widget` in the string and Markdown widget composables. - **What**: Delete both partial LiteGraph mocks so the tests exercise the real resolver. ## Review Focus The runtime function is unchanged. The tests keep their existing fake nodes and now use the real fallback graph-ID behavior. Validated with 32 focused tests, `pnpm typecheck`, lint, formatting, and `pnpm knip`. Part of [FE-2029](https://linear.app/comfyorg/issue/FE-2029/audit-and-reduce-importoriginal-usage-in-unit-tests). <details> <summary>Focused runtime measurements</summary> A cold focused comparison reduced process wall time from 4.25 s to 4.09 s, transform work from 3.18 s to 1.64 s, and import work from 4.86 s to 2.43 s. </details> Co-authored-by: Amp <amp@ampcode.com>
## Summary First PR of the Playwright coverage stack. ADR 0027 records how E2E coverage will be measured and how E2E tests are selected, backed by an analysis of the current suite. ## Changes - **What**: `docs/adr/0027-e2e-coverage-measurement-and-test-selection.md` (Proposed) plus index entry; `docs/testing/e2e-coverage-strategy.md` with the supporting numbers, gap inventory, redundant spec families, and reproduction steps, linked from `docs/testing/README.md`. Findings the ADR rests on (main @ `e05e929f`): - Reported E2E line coverage is 68.1%, but 457 `src/` files (~65k raw lines vs 70.5k measured) never load in any test and are missing from the lcov rather than counted at 0%. Over all of `src/` the figure is closer to 45 to 50%. - Only the `chromium` project sets `COLLECT_COVERAGE`. The 154 `cloud` tests are never credited, so workspace, billing, subscription, auth, and agent code look untested. - Vitest excludes `litegraph/src/canvas/**` and `widgets/**`, so E2E is the only measuring layer there (~49%). - `@smoke` is on 46 tests and no CI job runs it. There is no journey inventory. Decision in brief: collect coverage in every project, count never-loaded files at 0%, report per feature area through Codecov components (informational), keep `browser_tests/JOURNEYS.md` with one `@smoke` spec per journey run as a fast-fail project, and select E2E tests by journey and execution gap rather than by cross product. ## Review Focus - Whether the per-component, no-global-threshold policy in the Decision section matches how the team wants Codecov to behave. - The test-selection rules (items 7 to 11), since later PRs in the stack implement them. - The ADR is intentionally concise and the strategy doc carries the tables. The strategy doc can move to its own PR if the ADR should land alone. Follow-up PRs in the stack, in order: coverage in every Playwright project; never-loaded files at 0% plus a gap report; Codecov components; journey inventory and `smoke` project; gap-closing specs per area; consolidation of duplicated spec families. 🤖 Generated with [Claude Code](https://claude.com/claude-code) https://claude.ai/code/session_012csQMzkPkjHMAZPAP3wQZ7 --------- Co-authored-by: Amp <amp@ampcode.com> Co-authored-by: Claude <noreply@anthropic.com>
## Summary Adds a prerendered page for each of the 268 Workshop models when Workshop is enabled. Every page shows the model name, provider, description, tags, generated input form, and four related models. The form renders the five derived control kinds: text, select, number, toggle, and media upload. Generated defaults, required fields, numeric limits, accepted file types, and multiple-file roles are carried into the controls. Workshop remains absent from release builds. In Workshop-enabled labeled previews, the English-only catalog and model routes are indexable and represented consistently in the sitemap and hreflang metadata. ## Test plan - [x] Test all five field controls - [x] Test field defaults and related-model selection - [x] Astro and Vue typechecks pass - [x] Enabled build creates all 268 model routes - [x] Release build emits no Workshop routes - [x] Unknown model routes return 404 - [x] Enabled sitemap and hreflang targets are valid - [x] Knip and repository hooks pass
## Summary Adds TypeScript, Python, and raw HTTP examples beside each Workshop model form. The examples update as someone changes the form, so copied code includes the selected model and inputs. The TypeScript example uses `@comfyorg/sdk` and `comfy.models.run`. The Python example uses `comfy-sdk`. The HTTP example calls the current `/v2` Router endpoint. Media uploads are grouped into the Router `medias` list, including multiple files for roles that accept them. Payloads and model IDs are escaped for their target language. Regression coverage uses hostile model IDs containing both quote types, `$()`, and backticks; the HTTP snippet is executed against a stub `curl` to prove the endpoint remains one exact argument. Also links directly to the API key page. The Run button remains disabled until the sign-in and Router execution slice lands. ## Test plan - [x] Snapshot TypeScript, Python, and HTTP output - [x] Test form changes update the snippet - [x] Test JSON fields and single and multiple media roles - [x] Execute the HTTP snippet and verify apostrophes in payloads - [x] Preserve untrusted model IDs in all three snippet languages - [x] Full website unit suite passes - [x] Astro and Vue typechecks pass - [x] Release build emits no Workshop routes - [x] Enabled build creates all 268 model pages - [x] Knip and repository hooks pass
## Summary Two things, because the second one missed its own PR's merge window. **1. A Workshop section on the homepage.** Six hand-picked models spanning recognisable image generation, video, speech, music, and 3D capabilities, plus a link to the full catalog. The section reads the same build-time Workshop decision as the routes. Release builds emit neither the Workshop pages nor the homepage links. Workshop-enabled labeled previews emit the section, catalog, and model routes together. The featured list is explicit and editorial rather than a ranking, so regenerating the catalog cannot quietly change what the homepage leads with. If a partner retires a featured model, the card is skipped and a test fails so the list is corrected instead of silently shipping fewer cards. There is no `client:` directive on the section, so it renders to HTML and adds no hydrated island. **2. The idempotency-key fix that #16844 merged without.** #16844 entered the merge queue at `dbffc1094`, and the queue snapshotted it before this landed on the branch, so `main` currently carries the broken form. Carrying it here rather than opening a third PR for four files. `IDEMPOTENCY_KEY=$(uuidgen)` is not portable — `uuidgen` is absent from most Linux images, including CI's. The command failed and the request went out with an empty `Idempotency-Key`, which is the one value that header must never have, on calls that are billed. The suite stayed green because it asserted the URL and payload but never the key. The key is now minted in the page and rendered into the snippet as a literal, so re-running the copied block sends the key it sent the first time — a retry after a disconnect is a retry rather than a second charge. Three traps designed around: - **Minted in `onMounted`, not `setup`.** Setup also runs when Astro prerenders the island, so a key made there would be baked into static HTML and shared by every visitor — the Router would read one reader's run as a repeat of another's. Strictly worse than the empty header. Before hydration the snippet shows a visible placeholder, so a value copied early is rejected rather than silently colliding. - **Not `crypto.randomUUID()`.** It is restricted to secure contexts, so it is undefined on exactly the insecure LAN-IP origins the `legacy: true` clipboard fallback exists for. Built from `getRandomValues`, which has no such restriction. - **The key is a required parameter**, not defaulted, so no call site can quietly emit a keyless snippet. Reported by @DrJKL on #16844. ## Verification - [x] Release build (`WORKSHOP_IN_BUILD=0`): homepage contains zero `workshop` references, zero `/workshop` hrefs, no Workshop pages - [x] Deployed default (`VERCEL_ENV=preview`, no override): same, and the gate logs 269 pages removed - [x] Enabled build (`WORKSHOP_IN_BUILD=1`): 269 Workshop pages, 7 homepage links, section renders - [x] `WorkshopSection` appears in **0 JS chunks in every build** — no client directive, so the homepage pays nothing for it - [x] One gate on the path: `index.astro` calls `isWorkshopInBuild()` once, and no layout or shared header component references Workshop - [x] Featured IDs are covered against the packed catalog; a retired model is skipped, with a test that fails when the list goes stale - [x] Idempotency: the exact block executes twice and delivers the same non-empty key both times; runs with `PATH=` empty, proving no external binary; 50 keys distinct and v4-shaped; the component replaces its placeholder and two mounts differ - [x] Each of those four guards confirmed to **fail against a mutant that reintroduces the bug**, rather than trusted green - [x] 0 type errors, lint clean, 1260/1260 unit tests
## Summary Remove “Auto-run with limits” from the Agent run-permissions picker for V1. ## Changes Delete the third option from the options array. Its nested credit-limit input disappears with it. The production diff is six deleted lines; saved preferences, labels, API contracts, and backend behavior are unchanged. ## Review Focus Component tests assert that only two options are offered and cover saving Auto. Tests for editing the unavailable credit limit were removed; the original saved-mode tooltip tests remain unchanged. Validation: 86 Composer/store/REST-client tests passed, plus commit-hook formatting, lint, typecheck, and pre-push Knip. No new E2E test: availability is owned by this component and covered directly; no execution or API behavior changes. ## Screenshots Local visual validation remains pending because the existing Agent Storybook story fails to load an imported PNG as a JavaScript module.
…humbnails (#17147) 3D models in the media assets panel could not be downloaded or opened, and outputs of the Save3DAdvanced node family never got a thumbnail. Non-previewable assets (glb etc.) come back from the assets API without preview_url and without user_metadata.subfolder, so the /view URL built for them lacked subfolder=3d and the server returned 404. With the assets API enabled the file is now served by GET /api/assets/{id}/content, so the download, the 3D viewer dialog and the 3D card source resolve through one getAssetFileUrl instead of inferring a path. History-backed items keep using preview_url with the /view fallback. A card grouped per job carries the job id, so the resolver takes the asset id from its own allOutputs entry. Save3DAdvanced, SaveGaussianSplat and SavePointCloud have no model_file or image widget, so the modelReady thumbnail capture returned early and persistThumbnail never ran. Fall back to node.properties 'Last Time Model File' whenever the widget holds no non-blank string.
## Summary Replace the graph/core domain of #17222 plus local-instance removal from #17223 with global testing Pinia. ## Changes - 93 files including the identical four-file shared lifecycle prerequisite. Targets main and can merge independently. - Real store actions unless explicitly stubbed. Preserve scenario state and fixtures. - Enforcement and documentation remain deferred to #17223. ## Review Focus - All 89 domain paths are byte-identical to the corresponding files in [the immutable source](beb2e86). - `pnpm test:unit` with all 86 changed domain test paths plus three shared tests: 89 files passed, 1,243 tests passed and one expected failure. - Passed: `pnpm typecheck`, `pnpm typecheck:scripts`, scoped ESLint, type-aware Oxlint, cleanup audit, formatting, `pnpm knip`, and `git diff --check`. ESLint reports 69 warnings, zero errors. <details> <summary>Common lifecycle prerequisite</summary> - `vitest.setup.ts` - `scripts/testingPinia.test.ts` - `src/components/node/NodePreview.test.ts` - `src/renderer/extensions/vueNodes/components/LGraphNode.test.ts` </details> --------- Co-authored-by: Amp <amp@ampcode.com>
|
Review the following changes in direct dependencies. Learn more about Socket for GitHub.
|
d26a569 to
f789cda
Compare
|
Closing: this was a CI-only verification PR for the #14569 review, and its base branch has drifted far from main (+85k/-28k). The reproductions and CI runs stay linked here if that review picks up again. |
ELI5
When a workflow node contains a smaller workflow inside it (a subgraph), the outer box has to show you the picture that something deep inside is currently painting. Working out which inner thing a picture belongs to is the whole game, and the app labels those pictures with a number that is only unique inside one box — like two different offices both having a "Room 3". This PR doesn't fix anything; it is a set of recorded scenarios that show exactly where that labelling goes wrong, so reviewers can watch the failures happen on CI rather than take someone's word for it.
Summary
browser_tests/tests/subgraph/subgraphPreviewIdentity.spec.ts— 6 e2e scenarios, stacked on top of fix: decouple subgraph preview/output display from the promotion system #14569 (fix/decouple-subgraph-preview-outputs).browser_tests/assets/subgraphs/.src/lib/litegraph/src/subgraph/previewExposureRemap.repro.test.ts, a unit-level twin of scenario 4.Context
Review artifact for #14569. Each failing scenario corresponds to a point in the review comment on that PR; the passing ones exist so that a failure can't be blamed on a broken fixture.
Every fixture declares an explicit
previewExposuresarray on its hosts. That trips theproperties.previewExposures !== undefinedgate inautoExposeKnownPreviewNodes, so load-time auto-promotion cannot run and each scenario exercises the path it names instead of being satisfied by the exposure-driven path.Changes
The six scenarios
previewExposures: []technique really does isolate it from the promotion path.useAmbientSubgraphPreviewsbuilds its lookup key from the host's own local id, so a host at9:10looks up10:11while the frame is stored at9:10:11. Nested hosts get nothing from the ambient path.previewExposurespointing at a node that exists"3"vs interior["4"])deduplicateSubgraphNodeIdsrenumbers colliding interior ids but never patchespreviewExposures, so a promoted preview dies on reload. Predates #14569; that PR's ambient path incidentally masks the symptom for root-level hosts.Comfy.Graph.DeduplicateSubgraphNodeIdsoff, which the test sets.Supplementary
previewExposureRemap.repro.test.tsdrivesdeduplicateSubgraphNodeIdsdirectly and asserts the host exposure follows the renumbered interior node. Expected to fail withexpected "4", received "3"— the unit-level statement of scenario 4.Test plan
vue-tsc --project browser_tests/tsconfig.json,oxfmtandoxlint --type-awareclean on the added filessrc/file is a testplaywright-tests-chromium-sharded (13, 16)— 3 failed, 121 passed; the failures aresubgraphPreviewIdentity.spec.ts:45,:97,:129, i.e. scenarios 2, 4 and 5. Nothing else in the shard regressed.playwright-video-new-tests— 3 failed, 3 passed, with video artifacts of each failure.e2e-status— aggregate of the shard above.test(vitest) —previewExposureRemap.repro.test.ts, expected"4", received"3".